Link for the Competition: www.www.kaggle.com/competitions/train-your-own-stockfish-nnue

Link for my notebook: https://www.kaggle.com/code/hadriencr/train-your-own-stockfish-nnue

📰 Summary of the task

With the rise of NNUE for chess and shogi and machine learning getting more prominent in the digital cyberspace, I would like to introduce you to a competition that would challenge your inner creative monster and to create a NNUE model that is not an easy foe to overcome.

Goal: Use the dataset to train a neural network Chess Position Evaluator that can mimic stockfish's evaluation. The metric is the MAE to stockfish's evaluation (the (the reference used is: evalution +100 $\leftrightarrow$ an advantage of a pawn for the white side)

🗃️ Table of Contents

My approach to the competition

💭 Pipeline Design

⚙️ Model Design

model_1 architecture

🔄 Comparision of the optimizers

Finetuning

📋 What I learned from It

❓ What failed

✅ What worked

🔍 What could be explored further